def __init__(self, n_in, n_units, n_out):
super(MnistMLP, self).__init__(
l1=link_binary_linear.BinaryLinear(n_in, n_units),
b1=L.BatchNormalization(n_units),
l2=link_binary_linear.BinaryLinear(n_units, n_units),
b2=L.BatchNormalization(n_units),
l3=link_binary_linear.BinaryLinear(n_units, n_out),
b3=L.BatchNormalization(n_out),
)
self.train = True
def __init__(self):
super(BinaryConnectMnistMLP, self).__init__()
with self.init_scope():
self.bfc0 = link_binary_linear.BinaryLinear(784, 2048)
self.bn0 = L.BatchNormalization(2048)
self.bfc1 = link_binary_linear.BinaryLinear(2048, 2048)
self.bn1 = L.BatchNormalization(2048)
self.bfc2 = link_binary_linear.BinaryLinear(2048, 2048)
self.bn2 = L.BatchNormalization(2048)
self.bfc3 = link_binary_linear.BinaryLinear(2048, 10)
self.bn3 = L.BatchNormalization(10)
def __init__(self):
super(Cifar10CNN, self).__init__(
c1=link_binary_convolution.BinaryConvolution2D(3, 32, 5),
b1=L.BatchNormalization(32),
c2=link_binary_convolution.BinaryConvolution2D(32, 64, 5),
b2=L.BatchNormalization(64),
c3=link_binary_convolution.BinaryConvolution2D(64, 128, 5),
b3=L.BatchNormalization(128),
l1=link_binary_linear.BinaryLinear(128 * 20 * 20, 256),
b4=L.BatchNormalization(256),
l2=link_binary_linear.BinaryLinear(256, 10),
b5=L.BatchNormalization(10))
self.train = True
def __init__(self):
super(BinaryConnectMnistLeNet, self).__init__()
with self.init_scope():
self.bconv0 = link_binary_convolution.BinaryConvolution2D(1,
64,
ksize=5,
pad=0,
stride=1)
self.bn0 = L.BatchNormalization(64)
self.bconv1 = link_binary_convolution.BinaryConvolution2D(64,
64,
ksize=5,
pad=0,
stride=1)
self.bn1 = L.BatchNormalization(64)
self.bfc0 = link_binary_linear.BinaryLinear(1024, 512)
self.bn2 = L.BatchNormalization(512)
self.bfc1 = link_binary_linear.BinaryLinear(512, 10)
self.bn3 = L.BatchNormalization(10)
def __init__(self):
super(CNN, self).__init__(
conv0=L.Convolution2D(3,64,7, stride=2, pad=3, nobias=True),
b_conv0=L.BatchNormalization(64),
block0=RB.BlockStack(3,64,64,decre_ratio=4, kernel=(1,3,1), stride=(1,2,1), pad=(0,0,0),nobias=True),
b_block0=L.BatchNormalization(64),
conv1=IC.Convolution2D(64,128,3, stride=1, pad=1, nobias=True),
b_conv1=L.BatchNormalization(128),
block1=RB.BlockStack(3,128,128,decre_ratio=4, kernel=(1,3,1), stride=(1,2,1), pad=(0,0,0),nobias=True),
b_block1=L.BatchNormalization(128),
conv2=IC.Convolution2D(128,256,3, stride=1, pad=1, nobias=True),
b_conv2=L.BatchNormalization(256),
block2=RB.BlockStack(3,256,256,decre_ratio=4, kernel=(1,3,1), stride=(1,2,1), pad=(0,0,0),nobias=True),
b_block2=L.BatchNormalization(256),
conv3=IC.Convolution2D(256,512,3, stride=1, pad=1, nobias=True),
b_conv3=L.BatchNormalization(512),
block3=RB.BlockStack(3,512,512,decre_ratio=4, kernel=(1,3,1), stride=(1,2,1), pad=(0,0,0),nobias=True),
b_block3=L.BatchNormalization(512),
conv4=IC.Convolution2D(512,1024,3, stride=1, pad=1, nobias=True),
b_conv4=L.BatchNormalization(1024),
fc0=BL.BinaryLinear(4096,10),
b_dense0=L.BatchNormalization(10)
)
def __init__(self):
super(CNN, self).__init__(conv0=IC.Convolution2D(3,
64,
3,
stride=1,
pad=1,
nobias=True),
b0=LBN.BatchNormalization(64),
conv1=BC.Convolution2D(64,
128,
3,
stride=1,
pad=1,
nobias=True),
b1=LBN.BatchNormalization(128),
conv2=BC.Convolution2D(128,
128,
3,
stride=1,
pad=1,
nobias=True),
b2=LBN.BatchNormalization(128),
fc0=BL.BinaryLinear(128, 3),
b3=LBN.BatchNormalization(3))